ESymSolverStatus MumpsSolverInterface::SymbolicFactorization()
{
DBG_START_METH("MumpsSolverInterface::SymbolicFactorization",
dbg_verbosity);
DMUMPS_STRUC_C* mumps_data = (DMUMPS_STRUC_C*)mumps_ptr_;
if (HaveIpData()) {
IpData().TimingStats().LinearSystemSymbolicFactorization().Start();
}
mumps_data->job = 1;//symbolic ordering pass
//mumps_data->icntl[1] = 6;
//mumps_data->icntl[2] = 6;//QUIETLY!
//mumps_data->icntl[3] = 4;
mumps_data->icntl[5] = mumps_permuting_scaling_;
mumps_data->icntl[6] = mumps_pivot_order_;
mumps_data->icntl[7] = mumps_scaling_;
mumps_data->icntl[9] = 0;//no iterative refinement iterations
mumps_data->icntl[12] = 1;//avoid lapack bug, ensures proper inertia
mumps_data->icntl[13] = mem_percent_; //% memory to allocate over expected
mumps_data->cntl[0] = pivtol_; // Set pivot tolerance
dump_matrix(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Calling MUMPS-1 for symbolic factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
dmumps_c(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Done with MUMPS-1 for symbolic factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
int error = mumps_data->info[0];
const int& mumps_permuting_scaling_used = mumps_data->infog[22];
const int& mumps_pivot_order_used = mumps_data->infog[6];
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"MUMPS used permuting_scaling %d and pivot_order %d.\n",
mumps_permuting_scaling_used, mumps_pivot_order_used);
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
" scaling will be %d.\n",
mumps_data->icntl[7]);
if (HaveIpData()) {
IpData().TimingStats().LinearSystemSymbolicFactorization().End();
}
//return appropriat value
if (error == -6) {//system is singular
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) = %d matrix is singular.\n",error);
return SYMSOLVER_SINGULAR;
}
if (error < 0) {
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"Error=%d returned from MUMPS in Factorization.\n",
error);
return SYMSOLVER_FATAL_ERROR;
}
return SYMSOLVER_SUCCESS;
}
ESymSolverStatus MumpsSolverInterface::Factorization(
bool check_NegEVals, Index numberOfNegEVals)
{
DBG_START_METH("MumpsSolverInterface::Factorization", dbg_verbosity);
DMUMPS_STRUC_C* mumps_data = (DMUMPS_STRUC_C*)mumps_ptr_;
mumps_data->job = 2;//numerical factorization
dump_matrix(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Calling MUMPS-2 for numerical factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
dmumps_c(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Done with MUMPS-2 for numerical factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
int error = mumps_data->info[0];
//Check for errors
if (error == -8 || error == -9) {//not enough memory
const Index trycount_max = 20;
for (int trycount=0; trycount<trycount_max; trycount++) {
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) = %d and requires more memory, reallocating. Attempt %d\n",
error,trycount+1);
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA,
" Increasing icntl[13] from %d to ", mumps_data->icntl[13]);
double mem_percent = mumps_data->icntl[13];
mumps_data->icntl[13] = (Index)(2.0 * mem_percent);
Jnlst().Printf(J_WARNING, J_LINEAR_ALGEBRA, "%d.\n", mumps_data->icntl[13]);
dump_matrix(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Calling MUMPS-2 (repeated) for numerical factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
dmumps_c(mumps_data);
Jnlst().Printf(J_MOREDETAILED, J_LINEAR_ALGEBRA,
"Done with MUMPS-2 (repeated) for numerical factorization at cpu time %10.3f (wall %10.3f).\n", CpuTime(), WallclockTime());
error = mumps_data->info[0];
if (error != -8 && error != -9)
break;
}
if (error == -8 || error == -9) {
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"MUMPS was not able to obtain enough memory.\n");
return SYMSOLVER_FATAL_ERROR;
}
}
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"Number of doubles for MUMPS to hold factorization (INFO(9)) = %d\n",
mumps_data->info[8]);
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"Number of integers for MUMPS to hold factorization (INFO(10)) = %d\n",
mumps_data->info[9]);
if (error == -10) {//system is singular
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) = %d matrix is singular.\n",error);
return SYMSOLVER_SINGULAR;
}
negevals_ = mumps_data->infog[11];
if (error == -13) {
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) =%d - out of memory when trying to allocate %d %s.\nIn some cases it helps to decrease the value of the option \"mumps_mem_percent\".\n",
error, mumps_data->info[1] < 0 ? -mumps_data->info[1] : mumps_data->info[1], mumps_data->info[1] < 0 ? "MB" : "bytes");
return SYMSOLVER_FATAL_ERROR;
}
if (error < 0) {//some other error
Jnlst().Printf(J_ERROR, J_LINEAR_ALGEBRA,
"MUMPS returned INFO(1) =%d MUMPS failure.\n",
error);
return SYMSOLVER_FATAL_ERROR;
}
if (check_NegEVals && (numberOfNegEVals!=negevals_)) {
Jnlst().Printf(J_DETAILED, J_LINEAR_ALGEBRA,
"In MumpsSolverInterface::Factorization: negevals_ = %d, but numberOfNegEVals = %d\n",
negevals_, numberOfNegEVals);
return SYMSOLVER_WRONG_INERTIA;
}
return SYMSOLVER_SUCCESS;
}
int main(int argc, char *argv[])
{
int i, j, rtest, srcs, m, k, nerrs, r, err;
void *buf;
u8 g[TEST_SOURCES], g_tbls[TEST_SOURCES*32], src_in_err[TEST_SOURCES];
u8 *dest, *dest_ref, *temp_buff, *buffs[TEST_SOURCES];
u8 a[MMAX*KMAX], b[MMAX*KMAX], d[MMAX*KMAX];
u8 src_err_list[TEST_SOURCES], *recov[TEST_SOURCES];
int align, size;
unsigned char *efence_buffs[TEST_SOURCES];
unsigned int offset;
u8 *ubuffs[TEST_SOURCES];
u8 *udest_ptr;
printf("gf_vect_dot_prod_sse: %dx%d ", TEST_SOURCES, TEST_LEN);
mk_gf_field();
// Allocate the arrays
for(i=0; i<TEST_SOURCES; i++){
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
temp_buff = buf;
// Test of all zeros
for(i=0; i<TEST_SOURCES; i++)
memset(buffs[i], 0, TEST_LEN);
memset(dest, 0, TEST_LEN);
memset(temp_buff, 0, TEST_LEN);
memset(dest_ref, 0, TEST_LEN);
memset(g, 0, TEST_SOURCES);
for(i=0; i<TEST_SOURCES; i++)
gf_vect_mul_init(g[i], &g_tbls[i*32]);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref);
gf_vect_dot_prod_sse(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest);
if (0 != memcmp(dest_ref, dest, TEST_LEN)){
printf("Fail zero vect_dot_prod_sse test\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref, 25);
printf("dprod_sse:");
dump(dest, 25);;
return -1;
}
else
putchar('.');
// Rand data test
for(rtest=0; rtest<RANDOMS; rtest++){
for(i=0; i<TEST_SOURCES; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
for (i=0; i<TEST_SOURCES; i++)
g[i] = rand();
for(i=0; i<TEST_SOURCES; i++)
gf_vect_mul_init(g[i], &g_tbls[i*32]);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref);
gf_vect_dot_prod_sse(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest);
if (0 != memcmp(dest_ref, dest, TEST_LEN)){
printf("Fail rand vect_dot_prod_sse test 1\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref, 25);
printf("dprod_sse:");
dump(dest, 25);
return -1;
}
putchar('.');
}
// Rand data test with varied parameters
for(rtest=0; rtest < RANDOMS; rtest++){
for (srcs = TEST_SOURCES; srcs > 0; srcs--){
for(i=0; i<srcs; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
for (i=0; i<srcs; i++)
g[i] = rand();
for(i=0; i<srcs; i++)
gf_vect_mul_init(g[i], &g_tbls[i*32]);
gf_vect_dot_prod_base(TEST_LEN, srcs, &g_tbls[0], buffs, dest_ref);
gf_vect_dot_prod_sse(TEST_LEN, srcs, g_tbls, buffs, dest);
if (0 != memcmp(dest_ref, dest, TEST_LEN)){
printf("Fail rand vect_dot_prod_sse test 2\n");
dump_matrix(buffs, 5, srcs);
printf("dprod_base:");
dump(dest_ref, 5);
printf("dprod_sse:");
dump(dest, 5);
return -1;
}
putchar('.');
}
}
// Test erasure code using gf_vect_dot_prod
// Pick a first test
m = 9;
k = 5;
if (m > MMAX || k > KMAX)
return -1;
gf_gen_rs_matrix(a, m, k);
// Make random data
for(i=0; i<k; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
// Make parity vects
for (i=k; i<m; i++) {
for (j=0; j<k; j++)
gf_vect_mul_init(a[k*i+j], &g_tbls[j*32]);
#ifndef USEREF
gf_vect_dot_prod_sse(TEST_LEN,
k, g_tbls, buffs, buffs[i]);
#else
gf_vect_dot_prod_base(TEST_LEN,
k, &g_tbls[0], buffs, buffs[i]);
#endif
}
// Random buffers in erasure
memset(src_in_err, 0, TEST_SOURCES);
for (i=0, nerrs=0; i<k && nerrs<m-k; i++){
err = 1 & rand();
src_in_err[i] = err;
if (err)
src_err_list[nerrs++] = i;
}
// construct b by removing error rows
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r]) {
r++;
continue;
}
for(j=0; j<k; j++)
b[k*i+j] = a[k*r+j];
}
if (gf_invert_matrix((u8*)b, (u8*)d, k) < 0)
printf("BAD MATRIX\n");
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r]) {
r++;
continue;
}
recov[i] = buffs[r];
}
// Recover data
for(i=0; i<nerrs; i++){
for (j=0; j<k; j++)
gf_vect_mul_init(d[k*src_err_list[i]+j], &g_tbls[j*32]);
#ifndef USEREF
gf_vect_dot_prod_sse(TEST_LEN,
k, g_tbls, recov, temp_buff);
#else
gf_vect_dot_prod_base(TEST_LEN,
k, &g_tbls[0], recov, temp_buff);
#endif
if (0 != memcmp(temp_buff, buffs[src_err_list[i]],
TEST_LEN)){
printf("Fail error recovery (%d, %d, %d)\n", m, k, nerrs);
printf("recov %d:",src_err_list[i]);
dump(temp_buff, 25);
printf("orig :");
dump(buffs[src_err_list[i]],25);
return -1;
}
}
// Do more random tests
for (rtest = 0; rtest < RANDOMS; rtest++){
while ((m = (rand() % MMAX)) < 2);
while ((k = (rand() % KMAX)) >= m || k < 1);
if (m>MMAX || k>KMAX)
continue;
gf_gen_rs_matrix(a, m, k);
// Make random data
for(i=0; i<k; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
// Make parity vects
for (i=k; i<m; i++) {
for (j=0; j<k; j++)
gf_vect_mul_init(a[k*i+j], &g_tbls[j*32]);
#ifndef USEREF
gf_vect_dot_prod_sse(TEST_LEN, k, g_tbls, buffs, buffs[i]);
#else
gf_vect_dot_prod_base(TEST_LEN, k, &g_tbls[0], buffs, buffs[i]);
#endif
}
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
for (i=0, nerrs=0; i<k && nerrs<m-k; i++){
err = 1 & rand();
src_in_err[i] = err;
if (err)
src_err_list[nerrs++] = i;
}
if (nerrs == 0){ // should have at least one error
while ((err = (rand() % KMAX)) >= k) ;
src_err_list[nerrs++] = err;
src_in_err[err] = 1;
}
// construct b by removing error rows
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r]) {
r++;
continue;
}
for(j=0; j<k; j++)
b[k*i+j] = a[k*r+j];
}
if (gf_invert_matrix((u8*)b, (u8*)d, k) < 0)
printf("BAD MATRIX\n");
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r]) {
r++;
continue;
}
recov[i] = buffs[r];
}
// Recover data
for(i=0; i<nerrs; i++){
for (j=0; j<k; j++)
gf_vect_mul_init(d[k*src_err_list[i]+j], &g_tbls[j*32]);
#ifndef USEREF
gf_vect_dot_prod_sse(TEST_LEN, k, g_tbls, recov, temp_buff);
#else
gf_vect_dot_prod_base(TEST_LEN, k, &g_tbls[0], recov, temp_buff);
#endif
if (0 != memcmp(temp_buff, buffs[src_err_list[i]],
TEST_LEN)){
printf("Fail error recovery (%d, %d, %d) - ", m, k, nerrs);
printf(" - erase list = ");
for (i=0; i<nerrs; i++)
printf(" %d", src_err_list[i]);
printf("\na:\n");
dump_u8xu8((u8*)a, m, k);
printf("inv b:\n");
dump_u8xu8((u8*)d, k, k);
printf("orig data:\n");
dump_matrix(buffs, m, 25);
printf("orig :");
dump(buffs[src_err_list[i]],25);
printf("recov %d:",src_err_list[i]);
dump(temp_buff, 25);
return -1;
}
}
putchar('.');
}
// Run tests at end of buffer for Electric Fence
align = (LEN_ALIGN_CHK_B != 0) ? 1 : 16;
for(size=EFENCE_TEST_MIN_SIZE; size<=TEST_SIZE; size+=align){
for(i=0; i<TEST_SOURCES; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
for(i=0; i<TEST_SOURCES; i++) // Line up TEST_SIZE from end
efence_buffs[i] = buffs[i] + TEST_LEN - size;
for (i=0; i<TEST_SOURCES; i++)
g[i] = rand();
for(i=0; i<TEST_SOURCES; i++)
gf_vect_mul_init(g[i], &g_tbls[i*32]);
gf_vect_dot_prod_base(size, TEST_SOURCES, &g_tbls[0], efence_buffs, dest_ref);
gf_vect_dot_prod_sse(size, TEST_SOURCES, g_tbls, efence_buffs, dest);
if (0 != memcmp(dest_ref, dest, size)){
printf("Fail rand vect_dot_prod_sse test 3\n");
dump_matrix(efence_buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref, align);
printf("dprod_sse:");
dump(dest, align);
return -1;
}
putchar('.');
}
// Test rand ptr alignment if available
for(rtest=0; rtest<RANDOMS; rtest++){
size = (TEST_LEN - PTR_ALIGN_CHK_B) & ~15;
srcs = rand() % TEST_SOURCES;
if (srcs == 0)
continue;
offset = (PTR_ALIGN_CHK_B != 0) ? 1 : PTR_ALIGN_CHK_B;
// Add random offsets
for(i=0; i<srcs; i++)
ubuffs[i] = buffs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
udest_ptr = dest + (rand() & (PTR_ALIGN_CHK_B - offset));
memset(dest, 0, TEST_LEN); // zero pad to check write-over
for(i=0; i<srcs; i++)
for(j=0; j<size; j++)
ubuffs[i][j] = rand();
for (i=0; i<srcs; i++)
g[i] = rand();
for(i=0; i<srcs; i++)
gf_vect_mul_init(g[i], &g_tbls[i*32]);
gf_vect_dot_prod_base(size, srcs, &g_tbls[0], ubuffs, dest_ref);
gf_vect_dot_prod_sse(size, srcs, g_tbls, ubuffs, udest_ptr);
if (memcmp(dest_ref, udest_ptr, size)){
printf("Fail rand vect_dot_prod_sse test ualign srcs=%d\n", srcs);
dump_matrix(ubuffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref, 25);
printf("dprod_sse:");
dump(udest_ptr, 25);
return -1;
}
// Confirm that padding around dests is unchanged
memset(dest_ref, 0, PTR_ALIGN_CHK_B); // Make reference zero buff
offset = udest_ptr - dest;
if (memcmp(dest, dest_ref, offset)){
printf("Fail rand ualign pad start\n");
return -1;
}
if (memcmp(dest + offset + size, dest_ref, PTR_ALIGN_CHK_B - offset)){
printf("Fail rand ualign pad end\n");
return -1;
}
putchar('.');
}
// Test all size alignment
align = (LEN_ALIGN_CHK_B != 0) ? 1 : 16;
for(size=TEST_LEN; size>15; size-=align){
srcs = TEST_SOURCES;
for(i=0; i<srcs; i++)
for(j=0; j<size; j++)
buffs[i][j] = rand();
for (i=0; i<srcs; i++)
g[i] = rand();
for(i=0; i<srcs; i++)
gf_vect_mul_init(g[i], &g_tbls[i*32]);
gf_vect_dot_prod_base(size, srcs, &g_tbls[0], buffs, dest_ref);
gf_vect_dot_prod_sse(size, srcs, g_tbls, buffs, dest);
if (memcmp(dest_ref, dest, size)){
printf("Fail rand vect_dot_prod_sse test ualign len=%d\n", size);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref, 25);
printf("dprod_sse:");
dump(dest, 25);
return -1;
}
}
printf("done all: Pass\n");
return 0;
}
int main(int argc, char *argv[])
{
int i, j;
void *buf;
u8 g1[TEST_SOURCES], g2[TEST_SOURCES], g3[TEST_SOURCES];
u8 g_tbls[3 * TEST_SOURCES * 32], *dest_ptrs[3], *buffs[TEST_SOURCES];
u8 *dest1, *dest2, *dest3, *dest_ref1, *dest_ref2, *dest_ref3;
struct perf start, stop;
printf(xstr(FUNCTION_UNDER_TEST) ": %dx%d\n", TEST_SOURCES, TEST_LEN);
// Allocate the arrays
for (i = 0; i < TEST_SOURCES; i++) {
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest1 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest2 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest3 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref1 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref2 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref3 = buf;
dest_ptrs[0] = dest1;
dest_ptrs[1] = dest2;
dest_ptrs[2] = dest3;
// Performance test
for (i = 0; i < TEST_SOURCES; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
memset(dest1, 0, TEST_LEN);
memset(dest2, 0, TEST_LEN);
memset(dest_ref1, 0, TEST_LEN);
memset(dest_ref2, 0, TEST_LEN);
for (i = 0; i < TEST_SOURCES; i++) {
g1[i] = rand();
g2[i] = rand();
g3[i] = rand();
}
for (j = 0; j < TEST_SOURCES; j++) {
gf_vect_mul_init(g1[j], &g_tbls[j * 32]);
gf_vect_mul_init(g2[j], &g_tbls[(32 * TEST_SOURCES) + (j * 32)]);
gf_vect_mul_init(g3[j], &g_tbls[(64 * TEST_SOURCES) + (j * 32)]);
}
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[32 * TEST_SOURCES], buffs,
dest_ref2);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[64 * TEST_SOURCES], buffs,
dest_ref3);
#ifdef DO_REF_PERF
perf_start(&start);
for (i = 0; i < TEST_LOOPS / 100; i++) {
for (j = 0; j < TEST_SOURCES; j++) {
gf_vect_mul_init(g1[j], &g_tbls[j * 32]);
gf_vect_mul_init(g2[j], &g_tbls[(32 * TEST_SOURCES) + (j * 32)]);
gf_vect_mul_init(g3[j], &g_tbls[(64 * TEST_SOURCES) + (j * 32)]);
}
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[32 * TEST_SOURCES],
buffs, dest_ref2);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[64 * TEST_SOURCES],
buffs, dest_ref3);
}
perf_stop(&stop);
printf("gf_3vect_dot_prod_base" TEST_TYPE_STR ": ");
perf_print(stop, start, (long long)TEST_LEN * (TEST_SOURCES + 3) * i);
#endif
FUNCTION_UNDER_TEST(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ptrs);
perf_start(&start);
for (i = 0; i < TEST_LOOPS; i++) {
for (j = 0; j < TEST_SOURCES; j++) {
gf_vect_mul_init(g1[j], &g_tbls[j * 32]);
gf_vect_mul_init(g2[j], &g_tbls[(32 * TEST_SOURCES) + (j * 32)]);
gf_vect_mul_init(g3[j], &g_tbls[(64 * TEST_SOURCES) + (j * 32)]);
}
FUNCTION_UNDER_TEST(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ptrs);
}
perf_stop(&stop);
printf(xstr(FUNCTION_UNDER_TEST) TEST_TYPE_STR ": ");
perf_print(stop, start, (long long)TEST_LEN * (TEST_SOURCES + 3) * i);
if (0 != memcmp(dest_ref1, dest1, TEST_LEN)) {
printf("Fail perf " xstr(FUNCTION_UNDER_TEST) " test1\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_dut:");
dump(dest1, 25);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, TEST_LEN)) {
printf("Fail perf " xstr(FUNCTION_UNDER_TEST) " test2\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_dut:");
dump(dest2, 25);
return -1;
}
if (0 != memcmp(dest_ref3, dest3, TEST_LEN)) {
printf("Fail perf " xstr(FUNCTION_UNDER_TEST) " test3\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref3, 25);
printf("dprod_dut:");
dump(dest3, 25);
return -1;
}
printf("pass perf check\n");
return 0;
}
int main(int argc, char *argv[])
{
struct thread_data *threads;
struct thread_data *thread;
int i, ret, ch;
if (argc > 1)
{
if (strcmp(argv[1], "--help") == 0)
{
usage_error(argv[0]);
}
init_program_parameter(argc, argv);
}
program_parameter(argv[0]);
create_matrix(&matrix_a);
create_matrix(&matrix_b);
create_matrix(&matrix_c);
create_matrix(&matrix_d);
random_matrix(matrix_a);
random_matrix(matrix_b);
nonmal_matrix_multipy(matrix_a, matrix_b, matrix_d);
threads = (struct thread_data *)malloc(pthread_max * sizeof(struct thread_data));
if (threads == NULL)
{
unix_error("malloc threads failed");
}
cpu_online = sysconf(_SC_NPROCESSORS_CONF);
for(i = 0; i < pthread_max; i++)
{
thread = threads + i;
thread->index = i;
if ((ret = pthread_create(&thread->thread_id, NULL, thread_func, thread)) != 0)
{
posix_error(ret, "pthread_create failed");
}
}
for(i = 0; i < pthread_max; i++)
{
thread = threads + i;
if ((ret = pthread_join(thread->thread_id, NULL)) != 0)
{
posix_error(ret, "pthread_join failed");
}
}
if (matrix_equal(matrix_c, matrix_d) == 0)
{
unix_error("runtime error");
}
if (dump)
{
dump_matrix("matrix A", matrix_a);
dump_matrix("matrix B", matrix_b);
dump_matrix("matrix C", matrix_c);
dump_matrix("matrix D", matrix_d);
}
statistics(threads);
free_matrix(matrix_a);
free_matrix(matrix_b);
free_matrix(matrix_c);
free_matrix(matrix_d);
free(threads);
return 0;
}
int main(int argc, char *argv[])
{
int i, j, rtest, srcs;
void *buf;
u8 g1[TEST_SOURCES], g2[TEST_SOURCES], g3[TEST_SOURCES];
u8 g_tbls[3 * TEST_SOURCES * 32], *dest_ptrs[3], *buffs[TEST_SOURCES];
u8 *dest1, *dest2, *dest3, *dest_ref1, *dest_ref2, *dest_ref3;
int align, size;
unsigned char *efence_buffs[TEST_SOURCES];
unsigned int offset;
u8 *ubuffs[TEST_SOURCES];
u8 *udest_ptrs[3];
printf(xstr(FUNCTION_UNDER_TEST) "_test: %dx%d ", TEST_SOURCES, TEST_LEN);
// Allocate the arrays
for (i = 0; i < TEST_SOURCES; i++) {
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest1 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest2 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest3 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref1 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");;
return -1;
}
dest_ref2 = buf;
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref3 = buf;
dest_ptrs[0] = dest1;
dest_ptrs[1] = dest2;
dest_ptrs[2] = dest3;
// Test of all zeros
for (i = 0; i < TEST_SOURCES; i++)
memset(buffs[i], 0, TEST_LEN);
memset(dest1, 0, TEST_LEN);
memset(dest2, 0, TEST_LEN);
memset(dest3, 0, TEST_LEN);
memset(dest_ref1, 0, TEST_LEN);
memset(dest_ref2, 0, TEST_LEN);
memset(dest_ref3, 0, TEST_LEN);
memset(g1, 2, TEST_SOURCES);
memset(g2, 1, TEST_SOURCES);
memset(g3, 7, TEST_SOURCES);
for (i = 0; i < TEST_SOURCES; i++) {
gf_vect_mul_init(g1[i], &g_tbls[i * 32]);
gf_vect_mul_init(g2[i], &g_tbls[32 * TEST_SOURCES + i * 32]);
gf_vect_mul_init(g3[i], &g_tbls[64 * TEST_SOURCES + i * 32]);
}
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[32 * TEST_SOURCES], buffs,
dest_ref2);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[64 * TEST_SOURCES], buffs,
dest_ref3);
FUNCTION_UNDER_TEST(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ptrs);
if (0 != memcmp(dest_ref1, dest1, TEST_LEN)) {
printf("Fail zero" xstr(FUNCTION_UNDER_TEST) " test1\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_dut:");
dump(dest1, 25);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, TEST_LEN)) {
printf("Fail zero " xstr(FUNCTION_UNDER_TEST) " test2\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_dut:");
dump(dest2, 25);
return -1;
}
if (0 != memcmp(dest_ref3, dest3, TEST_LEN)) {
printf("Fail zero " xstr(FUNCTION_UNDER_TEST) " test3\n");
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref3, 25);
printf("dprod_dut:");
dump(dest3, 25);
return -1;
}
putchar('.');
// Rand data test
for (rtest = 0; rtest < RANDOMS; rtest++) {
for (i = 0; i < TEST_SOURCES; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
for (i = 0; i < TEST_SOURCES; i++) {
g1[i] = rand();
g2[i] = rand();
g3[i] = rand();
}
for (i = 0; i < TEST_SOURCES; i++) {
gf_vect_mul_init(g1[i], &g_tbls[i * 32]);
gf_vect_mul_init(g2[i], &g_tbls[(32 * TEST_SOURCES) + (i * 32)]);
gf_vect_mul_init(g3[i], &g_tbls[(64 * TEST_SOURCES) + (i * 32)]);
}
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[32 * TEST_SOURCES],
buffs, dest_ref2);
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[64 * TEST_SOURCES],
buffs, dest_ref3);
FUNCTION_UNDER_TEST(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ptrs);
if (0 != memcmp(dest_ref1, dest1, TEST_LEN)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test1 %d\n", rtest);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_dut:");
dump(dest1, 25);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, TEST_LEN)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test2 %d\n", rtest);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_dut:");
dump(dest2, 25);
return -1;
}
if (0 != memcmp(dest_ref3, dest3, TEST_LEN)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test3 %d\n", rtest);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref3, 25);
printf("dprod_dut:");
dump(dest3, 25);
return -1;
}
putchar('.');
}
// Rand data test with varied parameters
for (rtest = 0; rtest < RANDOMS; rtest++) {
for (srcs = TEST_SOURCES; srcs > 0; srcs--) {
for (i = 0; i < srcs; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
for (i = 0; i < srcs; i++) {
g1[i] = rand();
g2[i] = rand();
g3[i] = rand();
}
for (i = 0; i < srcs; i++) {
gf_vect_mul_init(g1[i], &g_tbls[i * 32]);
gf_vect_mul_init(g2[i], &g_tbls[(32 * srcs) + (i * 32)]);
gf_vect_mul_init(g3[i], &g_tbls[(64 * srcs) + (i * 32)]);
}
gf_vect_dot_prod_base(TEST_LEN, srcs, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(TEST_LEN, srcs, &g_tbls[32 * srcs], buffs,
dest_ref2);
gf_vect_dot_prod_base(TEST_LEN, srcs, &g_tbls[64 * srcs], buffs,
dest_ref3);
FUNCTION_UNDER_TEST(TEST_LEN, srcs, g_tbls, buffs, dest_ptrs);
if (0 != memcmp(dest_ref1, dest1, TEST_LEN)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST)
" test1 srcs=%d\n", srcs);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_dut:");
dump(dest1, 25);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, TEST_LEN)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST)
" test2 srcs=%d\n", srcs);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_dut:");
dump(dest2, 25);
return -1;
}
if (0 != memcmp(dest_ref3, dest3, TEST_LEN)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST)
" test3 srcs=%d\n", srcs);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref3, 25);
printf("dprod_dut:");
dump(dest3, 25);
return -1;
}
putchar('.');
}
}
// Run tests at end of buffer for Electric Fence
align = (LEN_ALIGN_CHK_B != 0) ? 1 : 16;
for (size = TEST_MIN_SIZE; size <= TEST_SIZE; size += align) {
for (i = 0; i < TEST_SOURCES; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
for (i = 0; i < TEST_SOURCES; i++) // Line up TEST_SIZE from end
efence_buffs[i] = buffs[i] + TEST_LEN - size;
for (i = 0; i < TEST_SOURCES; i++) {
g1[i] = rand();
g2[i] = rand();
g3[i] = rand();
}
for (i = 0; i < TEST_SOURCES; i++) {
gf_vect_mul_init(g1[i], &g_tbls[i * 32]);
gf_vect_mul_init(g2[i], &g_tbls[(32 * TEST_SOURCES) + (i * 32)]);
gf_vect_mul_init(g3[i], &g_tbls[(64 * TEST_SOURCES) + (i * 32)]);
}
gf_vect_dot_prod_base(size, TEST_SOURCES, &g_tbls[0], efence_buffs, dest_ref1);
gf_vect_dot_prod_base(size, TEST_SOURCES, &g_tbls[32 * TEST_SOURCES],
efence_buffs, dest_ref2);
gf_vect_dot_prod_base(size, TEST_SOURCES, &g_tbls[64 * TEST_SOURCES],
efence_buffs, dest_ref3);
FUNCTION_UNDER_TEST(size, TEST_SOURCES, g_tbls, efence_buffs, dest_ptrs);
if (0 != memcmp(dest_ref1, dest1, size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test1 %d\n", rtest);
dump_matrix(efence_buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, align);
printf("dprod_dut:");
dump(dest1, align);
return -1;
}
if (0 != memcmp(dest_ref2, dest2, size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test2 %d\n", rtest);
dump_matrix(efence_buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, align);
printf("dprod_dut:");
dump(dest2, align);
return -1;
}
if (0 != memcmp(dest_ref3, dest3, size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test3 %d\n", rtest);
dump_matrix(efence_buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref3, align);
printf("dprod_dut:");
dump(dest3, align);
return -1;
}
putchar('.');
}
// Test rand ptr alignment if available
for (rtest = 0; rtest < RANDOMS; rtest++) {
size = (TEST_LEN - PTR_ALIGN_CHK_B) & ~(TEST_MIN_SIZE - 1);
srcs = rand() % TEST_SOURCES;
if (srcs == 0)
continue;
offset = (PTR_ALIGN_CHK_B != 0) ? 1 : PTR_ALIGN_CHK_B;
// Add random offsets
for (i = 0; i < srcs; i++)
ubuffs[i] = buffs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
udest_ptrs[0] = dest1 + (rand() & (PTR_ALIGN_CHK_B - offset));
udest_ptrs[1] = dest2 + (rand() & (PTR_ALIGN_CHK_B - offset));
udest_ptrs[2] = dest3 + (rand() & (PTR_ALIGN_CHK_B - offset));
memset(dest1, 0, TEST_LEN); // zero pad to check write-over
memset(dest2, 0, TEST_LEN);
memset(dest3, 0, TEST_LEN);
for (i = 0; i < srcs; i++)
for (j = 0; j < size; j++)
ubuffs[i][j] = rand();
for (i = 0; i < srcs; i++) {
g1[i] = rand();
g2[i] = rand();
g3[i] = rand();
}
for (i = 0; i < srcs; i++) {
gf_vect_mul_init(g1[i], &g_tbls[i * 32]);
gf_vect_mul_init(g2[i], &g_tbls[(32 * srcs) + (i * 32)]);
gf_vect_mul_init(g3[i], &g_tbls[(64 * srcs) + (i * 32)]);
}
gf_vect_dot_prod_base(size, srcs, &g_tbls[0], ubuffs, dest_ref1);
gf_vect_dot_prod_base(size, srcs, &g_tbls[32 * srcs], ubuffs, dest_ref2);
gf_vect_dot_prod_base(size, srcs, &g_tbls[64 * srcs], ubuffs, dest_ref3);
FUNCTION_UNDER_TEST(size, srcs, g_tbls, ubuffs, udest_ptrs);
if (memcmp(dest_ref1, udest_ptrs[0], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test ualign srcs=%d\n",
srcs);
dump_matrix(ubuffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_dut:");
dump(udest_ptrs[0], 25);
return -1;
}
if (memcmp(dest_ref2, udest_ptrs[1], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test ualign srcs=%d\n",
srcs);
dump_matrix(ubuffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_dut:");
dump(udest_ptrs[1], 25);
return -1;
}
if (memcmp(dest_ref3, udest_ptrs[2], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test ualign srcs=%d\n",
srcs);
dump_matrix(ubuffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref3, 25);
printf("dprod_dut:");
dump(udest_ptrs[2], 25);
return -1;
}
// Confirm that padding around dests is unchanged
memset(dest_ref1, 0, PTR_ALIGN_CHK_B); // Make reference zero buff
offset = udest_ptrs[0] - dest1;
if (memcmp(dest1, dest_ref1, offset)) {
printf("Fail rand ualign pad1 start\n");
return -1;
}
if (memcmp(dest1 + offset + size, dest_ref1, PTR_ALIGN_CHK_B - offset)) {
printf("Fail rand ualign pad1 end\n");
return -1;
}
offset = udest_ptrs[1] - dest2;
if (memcmp(dest2, dest_ref1, offset)) {
printf("Fail rand ualign pad2 start\n");
return -1;
}
if (memcmp(dest2 + offset + size, dest_ref1, PTR_ALIGN_CHK_B - offset)) {
printf("Fail rand ualign pad2 end\n");
return -1;
}
offset = udest_ptrs[2] - dest3;
if (memcmp(dest3, dest_ref1, offset)) {
printf("Fail rand ualign pad3 start\n");
return -1;
}
if (memcmp(dest3 + offset + size, dest_ref1, PTR_ALIGN_CHK_B - offset)) {
printf("Fail rand ualign pad3 end\n");;
return -1;
}
putchar('.');
}
// Test all size alignment
align = (LEN_ALIGN_CHK_B != 0) ? 1 : 16;
for (size = TEST_LEN; size >= TEST_MIN_SIZE; size -= align) {
srcs = TEST_SOURCES;
for (i = 0; i < srcs; i++)
for (j = 0; j < size; j++)
buffs[i][j] = rand();
for (i = 0; i < srcs; i++) {
g1[i] = rand();
g2[i] = rand();
g3[i] = rand();
}
for (i = 0; i < srcs; i++) {
gf_vect_mul_init(g1[i], &g_tbls[i * 32]);
gf_vect_mul_init(g2[i], &g_tbls[(32 * srcs) + (i * 32)]);
gf_vect_mul_init(g3[i], &g_tbls[(64 * srcs) + (i * 32)]);
}
gf_vect_dot_prod_base(size, srcs, &g_tbls[0], buffs, dest_ref1);
gf_vect_dot_prod_base(size, srcs, &g_tbls[32 * srcs], buffs, dest_ref2);
gf_vect_dot_prod_base(size, srcs, &g_tbls[64 * srcs], buffs, dest_ref3);
FUNCTION_UNDER_TEST(size, srcs, g_tbls, buffs, dest_ptrs);
if (memcmp(dest_ref1, dest_ptrs[0], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test ualign len=%d\n",
size);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref1, 25);
printf("dprod_dut:");
dump(dest_ptrs[0], 25);
return -1;
}
if (memcmp(dest_ref2, dest_ptrs[1], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test ualign len=%d\n",
size);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref2, 25);
printf("dprod_dut:");
dump(dest_ptrs[1], 25);
return -1;
}
if (memcmp(dest_ref3, dest_ptrs[2], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test ualign len=%d\n",
size);
dump_matrix(buffs, 5, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref3, 25);
printf("dprod_dut:");
dump(dest_ptrs[2], 25);
return -1;
}
}
printf("Pass\n");
return 0;
}
void mexFunction
(
/* === Parameters ======================================================= */
int nlhs, /* number of left-hand sides */
mxArray *plhs [], /* left-hand side matrices */
int nrhs, /* number of right--hand sides */
const mxArray *prhs [] /* right-hand side matrices */
)
{
/* === Local variables ================================================== */
Long *perm ; /* column ordering of M and ordering of A */
Long *A ; /* row indices of input matrix A */
Long *p ; /* column pointers of input matrix A */
Long n_col ; /* number of columns of A */
Long n_row ; /* number of rows of A */
Long full ; /* TRUE if input matrix full, FALSE if sparse */
double knobs [COLAMD_KNOBS] ; /* colamd user-controllable parameters */
double *out_perm ; /* output permutation vector */
double *out_stats ; /* output stats vector */
double *in_knobs ; /* input knobs vector */
Long i ; /* loop counter */
mxArray *Ainput ; /* input matrix handle */
Long spumoni ; /* verbosity variable */
Long stats2 [COLAMD_STATS] ;/* stats for symamd */
Long *cp, *cp_end, result, nnz, col, length ;
Long *stats ;
stats = stats2 ;
/* === Check inputs ===================================================== */
if (nrhs < 1 || nrhs > 2 || nlhs < 0 || nlhs > 2)
{
mexErrMsgTxt (
"symamd: incorrect number of input and/or output arguments.") ;
}
if (nrhs != 2)
{
mexErrMsgTxt ("symamdtest: knobs are required") ;
}
/* for testing we require all 3 knobs */
if (mxGetNumberOfElements (prhs [1]) != 3)
{
mexErrMsgTxt ("symamdtest: must have all 3 knobs for testing") ;
}
/* === Get knobs ======================================================== */
colamd_l_set_defaults (knobs) ;
spumoni = 0 ;
/* check for user-passed knobs */
if (nrhs == 2)
{
in_knobs = mxGetPr (prhs [1]) ;
i = mxGetNumberOfElements (prhs [1]) ;
if (i > 0) knobs [COLAMD_DENSE_ROW] = in_knobs [0] ;
if (i > 1) spumoni = (Long) in_knobs [1] ;
}
/* print knob settings if spumoni is set */
if (spumoni)
{
mexPrintf ("\nsymamd version %d.%d, %s:\n",
COLAMD_MAIN_VERSION, COLAMD_SUB_VERSION, COLAMD_DATE) ;
if (knobs [COLAMD_DENSE_ROW] >= 0)
{
mexPrintf ("knobs(1): %g, rows/cols with > "
"max(16,%g*sqrt(size(A,2))) entries removed\n",
in_knobs [0], knobs [COLAMD_DENSE_ROW]) ;
}
else
{
mexPrintf ("knobs(1): %g, no dense rows removed\n", in_knobs [0]) ;
}
mexPrintf ("knobs(2): %g, statistics and knobs printed\n",
in_knobs [1]) ;
mexPrintf ("Testing %d\n", in_knobs [2]) ;
}
/* === If A is full, convert to a sparse matrix ========================= */
Ainput = (mxArray *) prhs [0] ;
if (mxGetNumberOfDimensions (Ainput) != 2)
{
mexErrMsgTxt ("symamd: input matrix must be 2-dimensional.") ;
}
full = !mxIsSparse (Ainput) ;
if (full)
{
mexCallMATLAB (1, &Ainput, 1, (mxArray **) prhs, "sparse") ;
}
/* === Allocate workspace for symamd ==================================== */
/* get size of matrix */
n_row = mxGetM (Ainput) ;
n_col = mxGetN (Ainput) ;
if (n_col != n_row)
{
mexErrMsgTxt ("symamd: matrix must be square.") ;
}
/* p = mxGetJc (Ainput) ; */
p = (Long *) mxCalloc (n_col+1, sizeof (Long)) ;
(void) memcpy (p, mxGetJc (Ainput), (n_col+1)*sizeof (Long)) ;
nnz = p [n_col] ;
if (spumoni > 0)
{
mexPrintf ("symamdtest: nnz %d\n", nnz) ;
}
/* A = mxGetIr (Ainput) ; */
A = (Long *) mxCalloc (nnz+1, sizeof (Long)) ;
(void) memcpy (A, mxGetIr (Ainput), nnz*sizeof (Long)) ;
perm = (Long *) mxCalloc (n_col+1, sizeof (Long)) ;
/* === Jumble matrix ======================================================== */
/*
knobs [2] FOR TESTING ONLY: Specifies how to jumble matrix
0 : No jumbling
1 : (no errors)
2 : Make first pointer non-zero
3 : Make column pointers not non-decreasing
4 : (no errors)
5 : Make row indices not strictly increasing
6 : Make a row index greater or equal to n_row
7 : Set A = NULL
8 : Set p = NULL
9 : Repeat row index
10: make row indices not sorted
11: jumble columns massively (note this changes
the pattern of the matrix A.)
12: Set stats = NULL
13: Make n_col less than zero
*/
/* jumble appropriately */
switch ((Long) in_knobs [2])
{
case 0 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: no errors expected\n") ;
}
result = 1 ; /* no errors */
break ;
case 1 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: no errors expected (1)\n") ;
}
result = 1 ;
break ;
case 2 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: p [0] nonzero\n") ;
}
result = 0 ; /* p [0] must be zero */
p [0] = 1 ;
break ;
case 3 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: negative length last column\n") ;
}
result = (n_col == 0) ; /* p must be monotonically inc. */
p [n_col] = p [0] ;
break ;
case 4 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: no errors expected (4)\n") ;
}
result = 1 ;
break ;
case 5 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: row index out of range (-1)\n") ;
}
if (nnz > 0) /* row index out of range */
{
result = 0 ;
A [nnz-1] = -1 ;
}
else
{
if (spumoni > 0)
{
mexPrintf ("Note: no row indices to put out of range\n") ;
}
result = 1 ;
}
break ;
case 6 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: row index out of range (ncol)\n") ;
}
if (nnz > 0) /* row index out of range */
{
result = 0 ;
A [nnz-1] = n_col ;
}
else
{
if (spumoni > 0)
{
mexPrintf ("Note: no row indices to put out of range\n") ;
}
result = 1 ;
}
break ;
case 7 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: A not present\n") ;
}
result = 0 ; /* A not present */
A = (Long *) NULL ;
break ;
case 8 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: p not present\n") ;
}
result = 0 ; /* p not present */
p = (Long *) NULL ;
break ;
case 9 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: duplicate row index\n") ;
}
result = 1 ; /* duplicate row index */
for (col = 0 ; col < n_col ; col++)
{
length = p [col+1] - p [col] ;
if (length > 1)
{
A [p [col+1]-2] = A [p [col+1] - 1] ;
if (spumoni > 0)
{
mexPrintf ("Made duplicate row %d in col %d\n",
A [p [col+1] - 1], col) ;
}
break ;
}
}
if (spumoni > 1)
{
dump_matrix (A, p, n_row, n_col, nnz, col+2) ;
}
break ;
case 10 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: unsorted column\n") ;
}
result = 1 ; /* jumbled columns */
for (col = 0 ; col < n_col ; col++)
{
length = p [col+1] - p [col] ;
if (length > 1)
{
i = A[p [col]] ;
A [p [col]] = A[p [col] + 1] ;
A [p [col] + 1] = i ;
if (spumoni > 0)
{
mexPrintf ("Unsorted column %d \n", col) ;
}
break ;
}
}
if (spumoni > 1)
{
dump_matrix (A, p, n_row, n_col, nnz, col+2) ;
}
break ;
case 11 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: massive jumbling\n") ;
}
result = 1 ; /* massive jumbling, but no errors */
srand (1) ;
for (i = 0 ; i < n_col ; i++)
{
cp = &A [p [i]] ;
cp_end = &A [p [i+1]] ;
while (cp < cp_end)
{
*cp++ = rand() % n_row ;
}
}
if (spumoni > 1)
{
dump_matrix (A, p, n_row, n_col, nnz, n_col) ;
}
break ;
case 12 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: stats not present\n") ;
}
result = 0 ; /* stats not present */
stats = (Long *) NULL ;
break ;
case 13 :
if (spumoni > 0)
{
mexPrintf ("symamdtest: ncol out of range\n") ;
}
result = 0 ; /* ncol out of range */
n_col = -1 ;
break ;
}
/* === Order the rows and columns of A (does not destroy A) ============= */
if (!symamd_l (n_col, A, p, perm, knobs, stats, &mxCalloc, &mxFree))
{
/* return p = -1 if colamd failed */
plhs [0] = mxCreateDoubleMatrix (1, 1, mxREAL) ;
out_perm = mxGetPr (plhs [0]) ;
out_perm [0] = -1 ;
mxFree (p) ;
mxFree (A) ;
if (spumoni > 0 || result)
{
symamd_l_report (stats) ;
}
if (result)
{
mexErrMsgTxt ("symamd should have returned TRUE\n") ;
}
return ;
/* mexErrMsgTxt ("symamd error!") ; */
}
if (!result)
{
symamd_l_report (stats) ;
mexErrMsgTxt ("symamd should have returned FALSE\n") ;
}
if (full)
{
mxDestroyArray (Ainput) ;
}
/* === Return the permutation vector ==================================== */
plhs [0] = mxCreateDoubleMatrix (1, n_col, mxREAL) ;
out_perm = mxGetPr (plhs [0]) ;
for (i = 0 ; i < n_col ; i++)
{
/* symamd is 0-based, but MATLAB expects this to be 1-based */
out_perm [i] = perm [i] + 1 ;
}
mxFree (perm) ;
/* === Return the stats vector ========================================== */
/* print stats if spumoni > 0 */
if (spumoni > 0)
{
symamd_l_report (stats) ;
}
if (nlhs == 2)
{
plhs [1] = mxCreateDoubleMatrix (1, COLAMD_STATS, mxREAL) ;
out_stats = mxGetPr (plhs [1]) ;
for (i = 0 ; i < COLAMD_STATS ; i++)
{
out_stats [i] = stats [i] ;
}
/* fix stats (5) and (6), for 1-based information on jumbled matrix. */
/* note that this correction doesn't occur if symamd returns FALSE */
out_stats [COLAMD_INFO1] ++ ;
out_stats [COLAMD_INFO2] ++ ;
}
}
/*
Compute the pseudoinverse of A and store it in A_plus
A+ = VS+U*
A m x n
V m x m
S n x n
U m x n
*/
int pseudoinverse(gsl_matrix *A, gsl_matrix *A_plus) {
int rc;
double temp;
int m = (int)(A->size1);
int n = (int)(A->size2);
gsl_vector *workspace = gsl_vector_alloc(n);
gsl_matrix *V = gsl_matrix_alloc(n, n);
gsl_matrix *S = gsl_matrix_alloc(n, n);
gsl_matrix *U_times_S_plus_trans = gsl_matrix_alloc(m, n);
gsl_vector *singular_values = gsl_vector_alloc(n);
int i;
double tolerance = 0;
rc = gsl_linalg_SV_decomp(A, V, singular_values, workspace); // NOTE: this does a thin SVD
// Now A = U
dump_matrix(A, "U.dat");
dump_matrix(V, "V.dat");
dump_vector(singular_values, "S.dat");
if (rc) {
// ERROR
return rc;
}
// compute tolerace as MAX(SIZE(A)) * NORM(A) * EPS(class(A))
// singular values within tolerance of 0 are considered to be 0
tolerance = MAX(m,n) * gsl_vector_get(singular_values, 0) * DBL_EPSILON;
//printf("tolerance = %g\n", tolerance);
// compute S+
// by taking reciprocal of nonzero entries
for (i = 0 ; i < n ; i++) {
temp = gsl_vector_get(singular_values, i);
if (temp < tolerance) {
break; // singular values are non-negative and form a non-increasing sequence
}
gsl_matrix_set(S, i, i, 1 / temp);
}
// now S = S+*
// compute US+*
rc = gsl_blas_dgemm(CblasNoTrans, CblasNoTrans, 1.0, A, S, 0.0, U_times_S_plus_trans);
// now A = US+*
if (rc) {
// ERROR
return rc;
}
// compute VS+U
rc = gsl_blas_dgemm(CblasNoTrans, CblasTrans, 1.0, V, U_times_S_plus_trans, 0.0, A_plus);
// now A_plus = VS+U
if (rc) {
// ERROR
return rc;
}
// cleanup
gsl_vector_free(workspace);
gsl_matrix_free(V);
gsl_matrix_free(S);
gsl_matrix_free(U_times_S_plus_trans);
gsl_vector_free(singular_values);
return 0;
}
int main(int argc, char *argv[])
{
int re = 0;
int i, j, p, rtest, m, k;
int nerrs, nsrcerrs;
void *buf;
unsigned int decode_index[MMAX];
unsigned char *temp_buffs[TEST_SOURCES], *buffs[TEST_SOURCES];
unsigned char *encode_matrix, *decode_matrix, *invert_matrix, *g_tbls;
unsigned char src_in_err[TEST_SOURCES], src_err_list[TEST_SOURCES];
unsigned char *recov[TEST_SOURCES];
int rows, align, size;
unsigned char *efence_buffs[TEST_SOURCES];
unsigned int offset;
u8 *ubuffs[TEST_SOURCES];
u8 *temp_ubuffs[TEST_SOURCES];
printf("erasure_code_test: %dx%d ", TEST_SOURCES, TEST_LEN);
srand(TEST_SEED);
// Allocate the arrays
for (i = 0; i < TEST_SOURCES; i++) {
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
for (i = 0; i < TEST_SOURCES; i++) {
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
temp_buffs[i] = buf;
}
// Test erasure code by encode and recovery
encode_matrix = malloc(MMAX * KMAX);
decode_matrix = malloc(MMAX * KMAX);
invert_matrix = malloc(MMAX * KMAX);
g_tbls = malloc(KMAX * TEST_SOURCES * 32);
if (encode_matrix == NULL || decode_matrix == NULL
|| invert_matrix == NULL || g_tbls == NULL) {
printf("Test failure! Error with malloc\n");
return -1;
}
// Pick a first test
m = 9;
k = 5;
if (m > MMAX || k > KMAX)
return -1;
// Make random data
for (i = 0; i < k; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
// Generate encode matrix encode_matrix
// The matrix generated by gf_gen_rs_matrix
// is not always invertable.
gf_gen_rs_matrix(encode_matrix, m, k);
// Generate g_tbls from encode matrix encode_matrix
ec_init_tables(k, m - k, &encode_matrix[k * k], g_tbls);
// Perform matrix dot_prod for EC encoding
// using g_tbls from encode matrix encode_matrix
ec_encode_data(TEST_LEN, k, m - k, g_tbls, buffs, &buffs[k]);
// Choose random buffers to be in erasure
memset(src_in_err, 0, TEST_SOURCES);
gen_err_list(src_err_list, src_in_err, &nerrs, &nsrcerrs, k, m);
// Generate decode matrix
re = gf_gen_decode_matrix(encode_matrix, decode_matrix,
invert_matrix, decode_index, src_err_list, src_in_err,
nerrs, nsrcerrs, k, m);
if (re != 0) {
printf("Fail to gf_gen_decode_matrix\n");
return -1;
}
// Pack recovery array as list of valid sources
// Its order must be the same as the order
// to generate matrix b in gf_gen_decode_matrix
for (i = 0; i < k; i++) {
recov[i] = buffs[decode_index[i]];
}
// Recover data
ec_init_tables(k, nerrs, decode_matrix, g_tbls);
ec_encode_data(TEST_LEN, k, nerrs, g_tbls, recov, &temp_buffs[k]);
for (i = 0; i < nerrs; i++) {
if (0 != memcmp(temp_buffs[k + i], buffs[src_err_list[i]], TEST_LEN)) {
printf("Fail error recovery (%d, %d, %d)\n", m, k, nerrs);
printf(" - erase list = ");
for (j = 0; j < nerrs; j++)
printf(" %d", src_err_list[j]);
printf(" - Index = ");
for (p = 0; p < k; p++)
printf(" %d", decode_index[p]);
printf("\nencode_matrix:\n");
dump_u8xu8((u8 *) encode_matrix, m, k);
printf("inv b:\n");
dump_u8xu8((u8 *) invert_matrix, k, k);
printf("\ndecode_matrix:\n");
dump_u8xu8((u8 *) decode_matrix, m, k);
printf("recov %d:", src_err_list[i]);
dump(temp_buffs[k + i], 25);
printf("orig :");
dump(buffs[src_err_list[i]], 25);
return -1;
}
}
// Pick a first test
m = 9;
k = 5;
if (m > MMAX || k > KMAX)
return -1;
// Make random data
for (i = 0; i < k; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
// The matrix generated by gf_gen_cauchy1_matrix
// is always invertable.
gf_gen_cauchy1_matrix(encode_matrix, m, k);
// Generate g_tbls from encode matrix encode_matrix
ec_init_tables(k, m - k, &encode_matrix[k * k], g_tbls);
// Perform matrix dot_prod for EC encoding
// using g_tbls from encode matrix encode_matrix
ec_encode_data(TEST_LEN, k, m - k, g_tbls, buffs, &buffs[k]);
// Choose random buffers to be in erasure
memset(src_in_err, 0, TEST_SOURCES);
gen_err_list(src_err_list, src_in_err, &nerrs, &nsrcerrs, k, m);
// Generate decode matrix
re = gf_gen_decode_matrix(encode_matrix, decode_matrix,
invert_matrix, decode_index, src_err_list, src_in_err,
nerrs, nsrcerrs, k, m);
if (re != 0) {
printf("Fail to gf_gen_decode_matrix\n");
return -1;
}
// Pack recovery array as list of valid sources
// Its order must be the same as the order
// to generate matrix b in gf_gen_decode_matrix
for (i = 0; i < k; i++) {
recov[i] = buffs[decode_index[i]];
}
// Recover data
ec_init_tables(k, nerrs, decode_matrix, g_tbls);
ec_encode_data(TEST_LEN, k, nerrs, g_tbls, recov, &temp_buffs[k]);
for (i = 0; i < nerrs; i++) {
if (0 != memcmp(temp_buffs[k + i], buffs[src_err_list[i]], TEST_LEN)) {
printf("Fail error recovery (%d, %d, %d)\n", m, k, nerrs);
printf(" - erase list = ");
for (j = 0; j < nerrs; j++)
printf(" %d", src_err_list[j]);
printf(" - Index = ");
for (p = 0; p < k; p++)
printf(" %d", decode_index[p]);
printf("\nencode_matrix:\n");
dump_u8xu8((u8 *) encode_matrix, m, k);
printf("inv b:\n");
dump_u8xu8((u8 *) invert_matrix, k, k);
printf("\ndecode_matrix:\n");
dump_u8xu8((u8 *) decode_matrix, m, k);
printf("recov %d:", src_err_list[i]);
dump(temp_buffs[k + i], 25);
printf("orig :");
dump(buffs[src_err_list[i]], 25);
return -1;
}
}
// Do more random tests
for (rtest = 0; rtest < RANDOMS; rtest++) {
while ((m = (rand() % MMAX)) < 2) ;
while ((k = (rand() % KMAX)) >= m || k < 1) ;
if (m > MMAX || k > KMAX)
continue;
// Make random data
for (i = 0; i < k; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
// The matrix generated by gf_gen_cauchy1_matrix
// is always invertable.
gf_gen_cauchy1_matrix(encode_matrix, m, k);
// Make parity vects
// Generate g_tbls from encode matrix a
ec_init_tables(k, m - k, &encode_matrix[k * k], g_tbls);
// Perform matrix dot_prod for EC encoding
// using g_tbls from encode matrix a
ec_encode_data(TEST_LEN, k, m - k, g_tbls, buffs, &buffs[k]);
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
gen_err_list(src_err_list, src_in_err, &nerrs, &nsrcerrs, k, m);
// Generate decode matrix
re = gf_gen_decode_matrix(encode_matrix, decode_matrix,
invert_matrix, decode_index, src_err_list,
src_in_err, nerrs, nsrcerrs, k, m);
if (re != 0) {
printf("Fail to gf_gen_decode_matrix\n");
return -1;
}
// Pack recovery array as list of valid sources
// Its order must be the same as the order
// to generate matrix b in gf_gen_decode_matrix
for (i = 0; i < k; i++) {
recov[i] = buffs[decode_index[i]];
}
// Recover data
ec_init_tables(k, nerrs, decode_matrix, g_tbls);
ec_encode_data(TEST_LEN, k, nerrs, g_tbls, recov, &temp_buffs[k]);
for (i = 0; i < nerrs; i++) {
if (0 != memcmp(temp_buffs[k + i], buffs[src_err_list[i]], TEST_LEN)) {
printf("Fail error recovery (%d, %d, %d) - ", m, k, nerrs);
printf(" - erase list = ");
for (j = 0; j < nerrs; j++)
printf(" %d", src_err_list[j]);
printf(" - Index = ");
for (p = 0; p < k; p++)
printf(" %d", decode_index[p]);
printf("\nencode_matrix:\n");
dump_u8xu8((u8 *) encode_matrix, m, k);
printf("inv b:\n");
dump_u8xu8((u8 *) invert_matrix, k, k);
printf("\ndecode_matrix:\n");
dump_u8xu8((u8 *) decode_matrix, m, k);
printf("orig data:\n");
dump_matrix(buffs, m, 25);
printf("orig :");
dump(buffs[src_err_list[i]], 25);
printf("recov %d:", src_err_list[i]);
dump(temp_buffs[k + i], 25);
return -1;
}
}
putchar('.');
}
// Run tests at end of buffer for Electric Fence
k = 16;
align = (LEN_ALIGN_CHK_B != 0) ? 1 : 16;
if (k > KMAX)
return -1;
for (rows = 1; rows <= 16; rows++) {
m = k + rows;
if (m > MMAX)
return -1;
// Make random data
for (i = 0; i < k; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
for (size = EFENCE_TEST_MIN_SIZE; size <= TEST_SIZE; size += align) {
for (i = 0; i < m; i++) { // Line up TEST_SIZE from end
efence_buffs[i] = buffs[i] + TEST_LEN - size;
}
// The matrix generated by gf_gen_cauchy1_matrix
// is always invertable.
gf_gen_cauchy1_matrix(encode_matrix, m, k);
// Make parity vects
// Generate g_tbls from encode matrix a
ec_init_tables(k, m - k, &encode_matrix[k * k], g_tbls);
// Perform matrix dot_prod for EC encoding
// using g_tbls from encode matrix a
ec_encode_data(size, k, m - k, g_tbls, efence_buffs, &efence_buffs[k]);
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
gen_err_list(src_err_list, src_in_err, &nerrs, &nsrcerrs, k, m);
// Generate decode matrix
re = gf_gen_decode_matrix(encode_matrix, decode_matrix,
invert_matrix, decode_index, src_err_list,
src_in_err, nerrs, nsrcerrs, k, m);
if (re != 0) {
printf("Fail to gf_gen_decode_matrix\n");
return -1;
}
// Pack recovery array as list of valid sources
// Its order must be the same as the order
// to generate matrix b in gf_gen_decode_matrix
for (i = 0; i < k; i++) {
recov[i] = efence_buffs[decode_index[i]];
}
// Recover data
ec_init_tables(k, nerrs, decode_matrix, g_tbls);
ec_encode_data(size, k, nerrs, g_tbls, recov, &temp_buffs[k]);
for (i = 0; i < nerrs; i++) {
if (0 !=
memcmp(temp_buffs[k + i], efence_buffs[src_err_list[i]],
size)) {
printf("Efence: Fail error recovery (%d, %d, %d)\n", m,
k, nerrs);
printf("size = %d\n", size);
printf("Test erase list = ");
for (j = 0; j < nerrs; j++)
printf(" %d", src_err_list[j]);
printf(" - Index = ");
for (p = 0; p < k; p++)
printf(" %d", decode_index[p]);
printf("\nencode_matrix:\n");
dump_u8xu8((u8 *) encode_matrix, m, k);
printf("inv b:\n");
dump_u8xu8((u8 *) invert_matrix, k, k);
printf("\ndecode_matrix:\n");
dump_u8xu8((u8 *) decode_matrix, m, k);
printf("recov %d:", src_err_list[i]);
dump(temp_buffs[k + i], align);
printf("orig :");
dump(efence_buffs[src_err_list[i]], align);
return -1;
}
}
}
}
// Test rand ptr alignment if available
for (rtest = 0; rtest < RANDOMS; rtest++) {
while ((m = (rand() % MMAX)) < 2) ;
while ((k = (rand() % KMAX)) >= m || k < 1) ;
if (m > MMAX || k > KMAX)
continue;
size = (TEST_LEN - PTR_ALIGN_CHK_B) & ~15;
offset = (PTR_ALIGN_CHK_B != 0) ? 1 : PTR_ALIGN_CHK_B;
// Add random offsets
for (i = 0; i < m; i++) {
memset(buffs[i], 0, TEST_LEN); // zero pad to check write-over
memset(temp_buffs[i], 0, TEST_LEN); // zero pad to check write-over
ubuffs[i] = buffs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
temp_ubuffs[i] = temp_buffs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
}
for (i = 0; i < k; i++)
for (j = 0; j < size; j++)
ubuffs[i][j] = rand();
// The matrix generated by gf_gen_cauchy1_matrix
// is always invertable.
gf_gen_cauchy1_matrix(encode_matrix, m, k);
// Make parity vects
// Generate g_tbls from encode matrix a
ec_init_tables(k, m - k, &encode_matrix[k * k], g_tbls);
// Perform matrix dot_prod for EC encoding
// using g_tbls from encode matrix a
ec_encode_data(size, k, m - k, g_tbls, ubuffs, &ubuffs[k]);
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
gen_err_list(src_err_list, src_in_err, &nerrs, &nsrcerrs, k, m);
// Generate decode matrix
re = gf_gen_decode_matrix(encode_matrix, decode_matrix,
invert_matrix, decode_index, src_err_list,
src_in_err, nerrs, nsrcerrs, k, m);
if (re != 0) {
printf("Fail to gf_gen_decode_matrix\n");
return -1;
}
// Pack recovery array as list of valid sources
// Its order must be the same as the order
// to generate matrix b in gf_gen_decode_matrix
for (i = 0; i < k; i++) {
recov[i] = ubuffs[decode_index[i]];
}
// Recover data
ec_init_tables(k, nerrs, decode_matrix, g_tbls);
ec_encode_data(size, k, nerrs, g_tbls, recov, &temp_ubuffs[k]);
for (i = 0; i < nerrs; i++) {
if (0 != memcmp(temp_ubuffs[k + i], ubuffs[src_err_list[i]], size)) {
printf("Fail error recovery (%d, %d, %d) - ", m, k, nerrs);
printf(" - erase list = ");
for (j = 0; j < nerrs; j++)
printf(" %d", src_err_list[j]);
printf(" - Index = ");
for (p = 0; p < k; p++)
printf(" %d", decode_index[p]);
printf("\nencode_matrix:\n");
dump_u8xu8((unsigned char *)encode_matrix, m, k);
printf("inv b:\n");
dump_u8xu8((unsigned char *)invert_matrix, k, k);
printf("\ndecode_matrix:\n");
dump_u8xu8((unsigned char *)decode_matrix, m, k);
printf("orig data:\n");
dump_matrix(ubuffs, m, 25);
printf("orig :");
dump(ubuffs[src_err_list[i]], 25);
printf("recov %d:", src_err_list[i]);
dump(temp_ubuffs[k + i], 25);
return -1;
}
}
// Confirm that padding around dests is unchanged
memset(temp_buffs[0], 0, PTR_ALIGN_CHK_B); // Make reference zero buff
for (i = 0; i < m; i++) {
offset = ubuffs[i] - buffs[i];
if (memcmp(buffs[i], temp_buffs[0], offset)) {
printf("Fail rand ualign encode pad start\n");
return -1;
}
if (memcmp
(buffs[i] + offset + size, temp_buffs[0],
PTR_ALIGN_CHK_B - offset)) {
printf("Fail rand ualign encode pad end\n");
return -1;
}
}
for (i = 0; i < nerrs; i++) {
offset = temp_ubuffs[k + i] - temp_buffs[k + i];
if (memcmp(temp_buffs[k + i], temp_buffs[0], offset)) {
printf("Fail rand ualign decode pad start\n");
return -1;
}
if (memcmp
(temp_buffs[k + i] + offset + size, temp_buffs[0],
PTR_ALIGN_CHK_B - offset)) {
printf("Fail rand ualign decode pad end\n");
return -1;
}
}
putchar('.');
}
// Test size alignment
align = (LEN_ALIGN_CHK_B != 0) ? 13 : 16;
for (size = TEST_LEN; size > 0; size -= align) {
while ((m = (rand() % MMAX)) < 2) ;
while ((k = (rand() % KMAX)) >= m || k < 1) ;
if (m > MMAX || k > KMAX)
continue;
for (i = 0; i < k; i++)
for (j = 0; j < size; j++)
buffs[i][j] = rand();
// The matrix generated by gf_gen_cauchy1_matrix
// is always invertable.
gf_gen_cauchy1_matrix(encode_matrix, m, k);
// Make parity vects
// Generate g_tbls from encode matrix a
ec_init_tables(k, m - k, &encode_matrix[k * k], g_tbls);
// Perform matrix dot_prod for EC encoding
// using g_tbls from encode matrix a
ec_encode_data(size, k, m - k, g_tbls, buffs, &buffs[k]);
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
gen_err_list(src_err_list, src_in_err, &nerrs, &nsrcerrs, k, m);
// Generate decode matrix
re = gf_gen_decode_matrix(encode_matrix, decode_matrix,
invert_matrix, decode_index, src_err_list,
src_in_err, nerrs, nsrcerrs, k, m);
if (re != 0) {
printf("Fail to gf_gen_decode_matrix\n");
return -1;
}
// Pack recovery array as list of valid sources
// Its order must be the same as the order
// to generate matrix b in gf_gen_decode_matrix
for (i = 0; i < k; i++) {
recov[i] = buffs[decode_index[i]];
}
// Recover data
ec_init_tables(k, nerrs, decode_matrix, g_tbls);
ec_encode_data(size, k, nerrs, g_tbls, recov, &temp_buffs[k]);
for (i = 0; i < nerrs; i++) {
if (0 != memcmp(temp_buffs[k + i], buffs[src_err_list[i]], size)) {
printf("Fail error recovery (%d, %d, %d) - ", m, k, nerrs);
printf(" - erase list = ");
for (j = 0; j < nerrs; j++)
printf(" %d", src_err_list[j]);
printf(" - Index = ");
for (p = 0; p < k; p++)
printf(" %d", decode_index[p]);
printf("\nencode_matrix:\n");
dump_u8xu8((unsigned char *)encode_matrix, m, k);
printf("inv b:\n");
dump_u8xu8((unsigned char *)invert_matrix, k, k);
printf("\ndecode_matrix:\n");
dump_u8xu8((unsigned char *)decode_matrix, m, k);
printf("orig data:\n");
dump_matrix(buffs, m, 25);
printf("orig :");
dump(buffs[src_err_list[i]], 25);
printf("recov %d:", src_err_list[i]);
dump(temp_buffs[k + i], 25);
return -1;
}
}
}
printf("done EC tests: Pass\n");
return 0;
}
int main(int argc, char *argv[])
{
int i, j, rtest, m, k, nerrs, r, err;
void *buf;
u8 *temp_buffs[TEST_SOURCES], *buffs[TEST_SOURCES];
u8 a[MMAX*KMAX], b[MMAX*KMAX], c[MMAX*KMAX], d[MMAX*KMAX];
u8 g_tbls[KMAX*TEST_SOURCES*32], src_in_err[TEST_SOURCES];
u8 src_err_list[TEST_SOURCES], *recov[TEST_SOURCES];
struct perf start, stop;
m = 32;
k = 28;
printf("erasure_code_sse_perf: %dx%d ",
m, (TEST_LEN(m)));
// Allocate the arrays
for(i=0; i<TEST_SOURCES; i++) {
if (posix_memalign(&buf, 64, TEST_LEN(m))) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
for (i=0; i<TEST_SOURCES; i++) {
if (posix_memalign(&buf, 64, TEST_LEN(m))) {
printf("alloc error: Fail");
return -1;
}
temp_buffs[i] = buf;
}
// Test erasure code by encode and recovery
// Pick a first test
if (m > MMAX || k > KMAX)
return -1;
// Make random data
for(i=0; i<k; i++)
for(j=0; j<(TEST_LEN(m)); j++)
buffs[i][j] = rand();
memset(src_in_err, 0, TEST_SOURCES);
srand(1);
for (i=0, nerrs=0; i<k && nerrs<m-k; i++) {
err = 1 & rand();
src_in_err[i] = err;
if (err)
src_err_list[nerrs++] = i;
}
if (nerrs == 0) { // should have at least one error
while ((err = (rand() % KMAX)) >= k) ;
src_err_list[nerrs++] = err;
src_in_err[err] = 1;
}
printf("Test erase list = ");
for (i=0; i<nerrs; i++)
printf(" %d", src_err_list[i]);
printf("\n");
perf_start(&start);
for (rtest = 0; rtest < TEST_LOOPS(m); rtest++) {
gf_gen_rs_matrix(a, m, k);
// Make parity vects
ec_init_tables(k, m-k, &a[k*k], g_tbls);
ec_encode_data_sse((TEST_LEN(m)),
k, m-k, g_tbls, buffs, &buffs[k]);
}
perf_stop(&stop);
printf("erasure_code_sse_encode" TEST_TYPE_STR ": ");
perf_print(stop,start,
(long long)(TEST_LEN(m))*(m)*rtest);
perf_start(&start);
for (rtest = 0; rtest < TEST_LOOPS(m); rtest++) {
// Construct b by removing error rows
for(i=0, r=0; i<k; i++, r++) {
while (src_in_err[r])
r++;
for(j=0; j<k; j++)
b[k*i+j] = a[k*r+j];
}
if (gf_invert_matrix(b, d, k) < 0) {
printf("BAD MATRIX\n");
return -1;
}
for(i=0, r=0; i<k; i++, r++) {
while (src_in_err[r])
r++;
recov[i] = buffs[r];
}
for(i=0; i<nerrs; i++) {
for(j=0; j<k; j++) {
c[k*i+j]=d[k*src_err_list[i]+j];
}
}
// Recover data
ec_init_tables(k, nerrs, c, g_tbls);
ec_encode_data_sse((TEST_LEN(m)),
k, nerrs, g_tbls, recov, &temp_buffs[k]);
}
perf_stop(&stop);
for(i=0; i<nerrs; i++) {
if (0 != memcmp(temp_buffs[k+i], buffs[src_err_list[i]],
(TEST_LEN(m)))) {
printf("Fail error recovery (%d, %d, %d) - ",
m, k, nerrs);
printf(" - erase list = ");
for (j=0; j<nerrs; j++)
printf(" %d", src_err_list[j]);
printf("\na:\n");
dump_u8xu8((u8*)a, m, k);
printf("inv b:\n");
dump_u8xu8((u8*)d, k, k);
printf("orig data:\n");
dump_matrix(buffs, m, 25);
printf("orig :");
dump(buffs[src_err_list[i]],25);
printf("recov %d:",src_err_list[i]);
dump(temp_buffs[k+i], 25);
return -1;
}
}
printf("erasure_code_sse_decode" TEST_TYPE_STR ": ");
perf_print(stop,start,
(long long)(TEST_LEN(m))*(k+nerrs)*rtest);
printf("done all: Pass\n");
return 0;
}
int main(int argc, char *argv[])
{
int i, j, rtest, srcs;
void *buf;
u8 gf[6][TEST_SOURCES];
u8 *g_tbls;
u8 *dest_ref[VECT];
u8 *dest_ptrs[VECT], *buffs[TEST_SOURCES];
int vector = VECT;
int align, size;
unsigned char *efence_buffs[TEST_SOURCES];
unsigned int offset;
u8 *ubuffs[TEST_SOURCES];
u8 *udest_ptrs[VECT];
printf("test" xstr(FUNCTION_UNDER_TEST) ": %dx%d ", TEST_SOURCES, TEST_LEN);
// Allocate the arrays
for (i = 0; i < TEST_SOURCES; i++) {
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
if (posix_memalign(&buf, 16, 2 * (vector * TEST_SOURCES * 32))) {
printf("alloc error: Fail");
return -1;
}
g_tbls = buf;
for (i = 0; i < vector; i++) {
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ptrs[i] = buf;
memset(dest_ptrs[i], 0, TEST_LEN);
}
for (i = 0; i < vector; i++) {
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
dest_ref[i] = buf;
memset(dest_ref[i], 0, TEST_LEN);
}
// Test of all zeros
for (i = 0; i < TEST_SOURCES; i++)
memset(buffs[i], 0, TEST_LEN);
switch (vector) {
case 6:
memset(gf[5], 0xe6, TEST_SOURCES);
case 5:
memset(gf[4], 4, TEST_SOURCES);
case 4:
memset(gf[3], 9, TEST_SOURCES);
case 3:
memset(gf[2], 7, TEST_SOURCES);
case 2:
memset(gf[1], 1, TEST_SOURCES);
case 1:
memset(gf[0], 2, TEST_SOURCES);
break;
default:
return -1;
}
for (i = 0; i < TEST_SOURCES; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
for (i = 0; i < vector; i++)
for (j = 0; j < TEST_SOURCES; j++) {
gf[i][j] = rand();
gf_vect_mul_init(gf[i][j], &g_tbls[i * (32 * TEST_SOURCES) + j * 32]);
}
for (i = 0; i < vector; i++)
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES, &g_tbls[i * 32 * TEST_SOURCES],
buffs, dest_ref[i]);
for (i = 0; i < vector; i++)
memset(dest_ptrs[i], 0, TEST_LEN);
for (i = 0; i < TEST_SOURCES; i++) {
#if (VECT == 1)
FUNCTION_UNDER_TEST(TEST_LEN, TEST_SOURCES, i, g_tbls, buffs[i], *dest_ptrs);
#else
FUNCTION_UNDER_TEST(TEST_LEN, TEST_SOURCES, i, g_tbls, buffs[i], dest_ptrs);
#endif
}
for (i = 0; i < vector; i++) {
if (0 != memcmp(dest_ref[i], dest_ptrs[i], TEST_LEN)) {
printf("Fail zero " xstr(FUNCTION_UNDER_TEST) " test%d\n", i);
dump_matrix(buffs, vector, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref[i], 25);
printf("dprod_dut:");
dump(dest_ptrs[i], 25);
return -1;
}
}
#if (VECT == 1)
REF_FUNCTION(TEST_LEN, TEST_SOURCES, g_tbls, buffs, *dest_ref);
#else
REF_FUNCTION(TEST_LEN, TEST_SOURCES, g_tbls, buffs, dest_ref);
#endif
for (i = 0; i < vector; i++) {
if (0 != memcmp(dest_ref[i], dest_ptrs[i], TEST_LEN)) {
printf("Fail zero " xstr(FUNCTION_UNDER_TEST) " test%d\n", i);
dump_matrix(buffs, vector, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref[i], 25);
printf("dprod_dut:");
dump(dest_ptrs[i], 25);
return -1;
}
}
putchar('.');
// Rand data test
for (rtest = 0; rtest < RANDOMS; rtest++) {
for (i = 0; i < TEST_SOURCES; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
for (i = 0; i < vector; i++)
for (j = 0; j < TEST_SOURCES; j++) {
gf[i][j] = rand();
gf_vect_mul_init(gf[i][j],
&g_tbls[i * (32 * TEST_SOURCES) + j * 32]);
}
for (i = 0; i < vector; i++)
gf_vect_dot_prod_base(TEST_LEN, TEST_SOURCES,
&g_tbls[i * 32 * TEST_SOURCES], buffs,
dest_ref[i]);
for (i = 0; i < vector; i++)
memset(dest_ptrs[i], 0, TEST_LEN);
for (i = 0; i < TEST_SOURCES; i++) {
#if (VECT == 1)
FUNCTION_UNDER_TEST(TEST_LEN, TEST_SOURCES, i, g_tbls, buffs[i],
*dest_ptrs);
#else
FUNCTION_UNDER_TEST(TEST_LEN, TEST_SOURCES, i, g_tbls, buffs[i],
dest_ptrs);
#endif
}
for (i = 0; i < vector; i++) {
if (0 != memcmp(dest_ref[i], dest_ptrs[i], TEST_LEN)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST) " test%d %d\n",
i, rtest);
dump_matrix(buffs, vector, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref[i], 25);
printf("dprod_dut:");
dump(dest_ptrs[i], 25);
return -1;
}
}
putchar('.');
}
// Rand data test with varied parameters
for (rtest = 0; rtest < RANDOMS; rtest++) {
for (srcs = TEST_SOURCES; srcs > 0; srcs--) {
for (i = 0; i < srcs; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
for (i = 0; i < vector; i++)
for (j = 0; j < srcs; j++) {
gf[i][j] = rand();
gf_vect_mul_init(gf[i][j],
&g_tbls[i * (32 * srcs) + j * 32]);
}
for (i = 0; i < vector; i++)
gf_vect_dot_prod_base(TEST_LEN, srcs, &g_tbls[i * 32 * srcs],
buffs, dest_ref[i]);
for (i = 0; i < vector; i++)
memset(dest_ptrs[i], 0, TEST_LEN);
for (i = 0; i < srcs; i++) {
#if (VECT == 1)
FUNCTION_UNDER_TEST(TEST_LEN, srcs, i, g_tbls, buffs[i],
*dest_ptrs);
#else
FUNCTION_UNDER_TEST(TEST_LEN, srcs, i, g_tbls, buffs[i],
dest_ptrs);
#endif
}
for (i = 0; i < vector; i++) {
if (0 != memcmp(dest_ref[i], dest_ptrs[i], TEST_LEN)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST)
" test%d srcs=%d\n", i, srcs);
dump_matrix(buffs, vector, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref[i], 25);
printf("dprod_dut:");
dump(dest_ptrs[i], 25);
return -1;
}
}
putchar('.');
}
}
// Run tests at end of buffer for Electric Fence
align = (LEN_ALIGN_CHK_B != 0) ? 1 : ALIGN_SIZE;
for (size = TEST_MIN_SIZE; size <= TEST_SIZE; size += align) {
for (i = 0; i < TEST_SOURCES; i++)
for (j = 0; j < TEST_LEN; j++)
buffs[i][j] = rand();
for (i = 0; i < TEST_SOURCES; i++) // Line up TEST_SIZE from end
efence_buffs[i] = buffs[i] + TEST_LEN - size;
for (i = 0; i < vector; i++)
for (j = 0; j < TEST_SOURCES; j++) {
gf[i][j] = rand();
gf_vect_mul_init(gf[i][j],
&g_tbls[i * (32 * TEST_SOURCES) + j * 32]);
}
for (i = 0; i < vector; i++)
gf_vect_dot_prod_base(size, TEST_SOURCES,
&g_tbls[i * 32 * TEST_SOURCES], efence_buffs,
dest_ref[i]);
for (i = 0; i < vector; i++)
memset(dest_ptrs[i], 0, size);
for (i = 0; i < TEST_SOURCES; i++) {
#if (VECT == 1)
FUNCTION_UNDER_TEST(size, TEST_SOURCES, i, g_tbls, efence_buffs[i],
*dest_ptrs);
#else
FUNCTION_UNDER_TEST(size, TEST_SOURCES, i, g_tbls, efence_buffs[i],
dest_ptrs);
#endif
}
for (i = 0; i < vector; i++) {
if (0 != memcmp(dest_ref[i], dest_ptrs[i], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST)
" test%d size=%d\n", i, size);
dump_matrix(buffs, vector, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref[i], TEST_MIN_SIZE + align);
printf("dprod_dut:");
dump(dest_ptrs[i], TEST_MIN_SIZE + align);
return -1;
}
}
putchar('.');
}
// Test rand ptr alignment if available
for (rtest = 0; rtest < RANDOMS; rtest++) {
size = (TEST_LEN - PTR_ALIGN_CHK_B) & ~(TEST_MIN_SIZE - 1);
srcs = rand() % TEST_SOURCES;
if (srcs == 0)
continue;
offset = (PTR_ALIGN_CHK_B != 0) ? 1 : PTR_ALIGN_CHK_B;
// Add random offsets
for (i = 0; i < srcs; i++)
ubuffs[i] = buffs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
for (i = 0; i < vector; i++) {
udest_ptrs[i] = dest_ptrs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
memset(dest_ptrs[i], 0, TEST_LEN); // zero pad to check write-over
}
for (i = 0; i < srcs; i++)
for (j = 0; j < size; j++)
ubuffs[i][j] = rand();
for (i = 0; i < vector; i++)
for (j = 0; j < srcs; j++) {
gf[i][j] = rand();
gf_vect_mul_init(gf[i][j], &g_tbls[i * (32 * srcs) + j * 32]);
}
for (i = 0; i < vector; i++)
gf_vect_dot_prod_base(size, srcs, &g_tbls[i * 32 * srcs], ubuffs,
dest_ref[i]);
for (i = 0; i < srcs; i++) {
#if (VECT == 1)
FUNCTION_UNDER_TEST(size, srcs, i, g_tbls, ubuffs[i], *udest_ptrs);
#else
FUNCTION_UNDER_TEST(size, srcs, i, g_tbls, ubuffs[i], udest_ptrs);
#endif
}
for (i = 0; i < vector; i++) {
if (0 != memcmp(dest_ref[i], udest_ptrs[i], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST)
" test%d ualign srcs=%d\n", i, srcs);
dump_matrix(buffs, vector, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref[i], 25);
printf("dprod_dut:");
dump(udest_ptrs[i], 25);
return -1;
}
}
// Confirm that padding around dests is unchanged
memset(dest_ref[0], 0, PTR_ALIGN_CHK_B); // Make reference zero buff
for (i = 0; i < vector; i++) {
offset = udest_ptrs[i] - dest_ptrs[i];
if (memcmp(dest_ptrs[i], dest_ref[0], offset)) {
printf("Fail rand ualign pad1 start\n");
return -1;
}
if (memcmp
(dest_ptrs[i] + offset + size, dest_ref[0],
PTR_ALIGN_CHK_B - offset)) {
printf("Fail rand ualign pad1 end\n");
return -1;
}
}
putchar('.');
}
// Test all size alignment
align = (LEN_ALIGN_CHK_B != 0) ? 1 : ALIGN_SIZE;
for (size = TEST_LEN; size >= TEST_MIN_SIZE; size -= align) {
for (i = 0; i < TEST_SOURCES; i++)
for (j = 0; j < size; j++)
buffs[i][j] = rand();
for (i = 0; i < vector; i++) {
for (j = 0; j < TEST_SOURCES; j++) {
gf[i][j] = rand();
gf_vect_mul_init(gf[i][j],
&g_tbls[i * (32 * TEST_SOURCES) + j * 32]);
}
memset(dest_ptrs[i], 0, TEST_LEN); // zero pad to check write-over
}
for (i = 0; i < vector; i++)
gf_vect_dot_prod_base(size, TEST_SOURCES,
&g_tbls[i * 32 * TEST_SOURCES], buffs,
dest_ref[i]);
for (i = 0; i < TEST_SOURCES; i++) {
#if (VECT == 1)
FUNCTION_UNDER_TEST(size, TEST_SOURCES, i, g_tbls, buffs[i],
*dest_ptrs);
#else
FUNCTION_UNDER_TEST(size, TEST_SOURCES, i, g_tbls, buffs[i],
dest_ptrs);
#endif
}
for (i = 0; i < vector; i++) {
if (0 != memcmp(dest_ref[i], dest_ptrs[i], size)) {
printf("Fail rand " xstr(FUNCTION_UNDER_TEST)
" test%d ualign len=%d\n", i, size);
dump_matrix(buffs, vector, TEST_SOURCES);
printf("dprod_base:");
dump(dest_ref[i], 25);
printf("dprod_dut:");
dump(dest_ptrs[i], 25);
return -1;
}
}
putchar('.');
}
printf("Pass\n");
return 0;
}
int matrix_invert(FILE *fp, int n, double **a)
{
int i, j, m, lda, *ipiv, lwork, info;
double **test = NULL, **id, *work;
#ifdef DEBUG_MATRIX
if (fp)
{
fprintf(fp, "Inverting %d square matrix\n", n);
test = alloc_matrix(n, n);
for (i = 0; (i < n); i++)
{
for (j = 0; (j < n); j++)
{
test[i][j] = a[i][j];
}
}
dump_matrix(fp, "before inversion", n, a);
}
#endif
snew(ipiv, n);
lwork = n*n;
snew(work, lwork);
m = lda = n;
info = 0;
F77_FUNC(dgetrf, DGETRF) (&n, &m, a[0], &lda, ipiv, &info);
#ifdef DEBUG_MATRIX
if (fp)
{
dump_matrix(fp, "after dgetrf", n, a);
}
#endif
if (info != 0)
{
return info;
}
F77_FUNC(dgetri, DGETRI) (&n, a[0], &lda, ipiv, work, &lwork, &info);
#ifdef DEBUG_MATRIX
if (fp)
{
dump_matrix(fp, "after dgetri", n, a);
}
#endif
if (info != 0)
{
return info;
}
#ifdef DEBUG_MATRIX
if (fp)
{
id = alloc_matrix(n, n);
matrix_multiply(fp, n, n, test, a, id);
dump_matrix(fp, "And here is the product of A and Ainv", n, id);
free_matrix(id);
free_matrix(test);
}
#endif
sfree(ipiv);
sfree(work);
return 0;
}
int main(int argc, char *argv[])
{
int i, j, rtest, m, k, nerrs, r, err;
void *buf;
unsigned char *temp_buffs[TEST_SOURCES], *buffs[TEST_SOURCES], *a, *b, *c, *d, *g_tbls;
unsigned char src_in_err[TEST_SOURCES], src_err_list[TEST_SOURCES];
unsigned char *recov[TEST_SOURCES];
int rows, align, size;
unsigned char *efence_buffs[TEST_SOURCES];
unsigned int offset;
u8 *ubuffs[TEST_SOURCES];
u8 *temp_ubuffs[TEST_SOURCES];
printf("erasure_code_sse_test: %dx%d ", TEST_SOURCES, TEST_LEN);
// Allocate the arrays
for(i=0; i<TEST_SOURCES; i++){
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
buffs[i] = buf;
}
for(i=0; i<TEST_SOURCES; i++){
if (posix_memalign(&buf, 64, TEST_LEN)) {
printf("alloc error: Fail");
return -1;
}
temp_buffs[i] = buf;
}
// Test erasure code by encode and recovery
a = malloc(MMAX*KMAX);
b = malloc(MMAX*KMAX);
c = malloc(MMAX*KMAX);
d = malloc(MMAX*KMAX);
g_tbls = malloc(KMAX*TEST_SOURCES*32);
if (a == NULL || b == NULL || c == NULL || d == NULL || g_tbls == NULL) {
printf("Test failure! Error with malloc\n");
return -1;
}
// Pick a first test
m = 9;
k = 5;
if (m > MMAX || k > KMAX)
return -1;
// Make random data
for(i=0; i<k; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
gf_gen_rs_matrix(a, m, k);
ec_init_tables(k, m-k, &a[k*k], g_tbls);
ec_encode_data_sse(TEST_LEN, k, m-k, g_tbls, buffs, &buffs[k]);
// Choose random buffers to be in erasure
memset(src_in_err, 0, TEST_SOURCES);
for (i=0, nerrs=0; i<k && nerrs<m-k; i++){
err = 1 & rand();
src_in_err[i] = err;
if (err)
src_err_list[nerrs++] = i;
}
// Construct matrix b by removing error rows
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
for(j=0; j<k; j++)
b[k*i+j] = a[k*r+j];
}
// Generate decode matrix d as matrix inverse of b
if (gf_invert_matrix(b, d, k) < 0){
printf("BAD MATRIX\n");
return -1;
}
// Pack recovery array as list of valid sources
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
recov[i] = buffs[r];
}
for(i=0; i<nerrs; i++){
for(j=0; j<k; j++){
c[k*i+j]=d[k*src_err_list[i]+j];
}
}
// Recover data
ec_init_tables(k, nerrs, c, g_tbls);
ec_encode_data(TEST_LEN,
k, nerrs, g_tbls, recov, &temp_buffs[k]);
for(i=0; i<nerrs; i++){
if (0 != memcmp(temp_buffs[k+i], buffs[src_err_list[i]], TEST_LEN)){
printf("Fail error recovery (%d, %d, %d)\n", m, k, nerrs);
printf("recov %d:",src_err_list[i]);
dump(temp_buffs[k+i], 25);
printf("orig :");
dump(buffs[src_err_list[i]],25);
return -1;
}
}
// Do more random tests
for (rtest = 0; rtest < RANDOMS; rtest++){
while ((m = (rand() % MMAX)) < 2);
while ((k = (rand() % KMAX)) >= m || k < 1);
if (m>MMAX || k>KMAX)
continue;
// Make random data
for(i=0; i<k; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
gf_gen_rs_matrix(a, m, k);
// Make parity vects
ec_init_tables(k, m-k, &a[k*k], g_tbls);
ec_encode_data_sse(TEST_LEN, k, m-k, g_tbls, buffs, &buffs[k]);
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
for (i=0, nerrs=0; i<k && nerrs<m-k; i++){
err = 1 & rand();
src_in_err[i] = err;
if (err)
src_err_list[nerrs++] = i;
}
if (nerrs == 0){ // should have at least one error
while ((err = (rand() % KMAX)) >= k) ;
src_err_list[nerrs++] = err;
src_in_err[err] = 1;
}
// Construct b by removing error rows
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
for(j=0; j<k; j++)
b[k*i+j] = a[k*r+j];
}
if (gf_invert_matrix(b, d, k) < 0){
printf("BAD MATRIX\n");
return -1;
}
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
recov[i] = buffs[r];
}
for(i=0; i<nerrs; i++){
for(j=0; j<k; j++){
c[k*i+j]=d[k*src_err_list[i]+j];
}
}
// Recover data
ec_init_tables(k, nerrs, c, g_tbls);
ec_encode_data(TEST_LEN, k, nerrs, g_tbls, recov, &temp_buffs[k]);
for(i=0; i<nerrs; i++){
if (0 != memcmp(temp_buffs[k+i], buffs[src_err_list[i]], TEST_LEN)){
printf("Fail error recovery (%d, %d, %d) - ", m, k, nerrs);
printf(" - erase list = ");
for (i=0; i<nerrs; i++)
printf(" %d", src_err_list[i]);
printf("\na:\n");
dump_u8xu8((unsigned char*)a, m, k);
printf("inv b:\n");
dump_u8xu8((unsigned char*)d, k, k);
printf("orig data:\n");
dump_matrix(buffs, m, 25);
printf("orig :");
dump(buffs[src_err_list[i]],25);
printf("recov %d:",src_err_list[i]);
dump(temp_buffs[k+i], 25);
return -1;
}
}
putchar('.');
}
// Run tests at end of buffer for Electric Fence
k = 16;
align = (LEN_ALIGN_CHK_B != 0) ? 1 : 16;
if (k > KMAX)
return -1;
for(rows=1; rows<=16; rows++){
m = k+rows;
if (m > MMAX)
return -1;
// Make random data
for(i=0; i<k; i++)
for(j=0; j<TEST_LEN; j++)
buffs[i][j] = rand();
for(size=EFENCE_TEST_MIN_SIZE; size<=TEST_SIZE; size+=align){
for(i=0; i<m; i++) // Line up TEST_SIZE from end
efence_buffs[i] = buffs[i] + TEST_LEN - size;
gf_gen_rs_matrix(a, m, k);
ec_init_tables(k, m-k, &a[k*k], g_tbls);
ec_encode_data_sse(size, k, m-k, g_tbls, efence_buffs, &efence_buffs[k]);
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
for (i=0, nerrs=0; i<k && nerrs<m-k; i++){
err = 1 & rand();
src_in_err[i] = err;
if (err)
src_err_list[nerrs++] = i;
}
if (nerrs == 0){ // should have at least one error
while ((err = (rand() % KMAX)) >= k) ;
src_err_list[nerrs++] = err;
src_in_err[err] = 1;
}
// Construct b by removing error rows
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
for(j=0; j<k; j++)
b[k*i+j] = a[k*r+j];
}
// Generate decode matrix d as matrix inverse of b
if (gf_invert_matrix(b, d, k) < 0){
printf("BAD MATRIX\n");
return -1;
}
// Pack recovery array as list of valid sources
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
recov[i] = efence_buffs[r];
}
for(i=0; i<nerrs; i++){
for(j=0; j<k; j++){
c[k*i+j]=d[k*src_err_list[i]+j];
}
}
// Recover data
ec_init_tables(k, nerrs, c, g_tbls);
ec_encode_data(size, k, nerrs, g_tbls, recov, &temp_buffs[k]);
for(i=0; i<nerrs; i++){
if (0 != memcmp(temp_buffs[k+i], efence_buffs[src_err_list[i]], size)){
printf("Efence: Fail error recovery (%d, %d, %d)\n", m, k, nerrs);
printf("Test erase list = ");
for (i=0; i<nerrs; i++)
printf(" %d", src_err_list[i]);
printf("\n");
printf("recov %d:",src_err_list[i]);
dump(temp_buffs[k+i], align);
printf("orig :");
dump(efence_buffs[src_err_list[i]],align);
return -1;
}
}
}
}
// Test rand ptr alignment if available
for(rtest=0; rtest<RANDOMS; rtest++){
while ((m = (rand() % MMAX)) < 2);
while ((k = (rand() % KMAX)) >= m || k < 1);
if (m>MMAX || k>KMAX)
continue;
size = (TEST_LEN - PTR_ALIGN_CHK_B) & ~15;
offset = (PTR_ALIGN_CHK_B != 0) ? 1 : PTR_ALIGN_CHK_B;
// Add random offsets
for(i=0; i<m; i++) {
memset(buffs[i], 0, TEST_LEN); // zero pad to check write-over
memset(temp_buffs[i], 0, TEST_LEN); // zero pad to check write-over
ubuffs[i] = buffs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
temp_ubuffs[i] = temp_buffs[i] + (rand() & (PTR_ALIGN_CHK_B - offset));
}
for(i=0; i<k; i++)
for(j=0; j<size; j++)
ubuffs[i][j] = rand();
gf_gen_rs_matrix(a, m, k);
// Make parity vects
ec_init_tables(k, m-k, &a[k*k], g_tbls);
ec_encode_data_sse(size, k, m-k, g_tbls, ubuffs, &ubuffs[k]);
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
for (i=0, nerrs=0; i<k && nerrs<m-k; i++){
err = 1 & rand();
src_in_err[i] = err;
if (err)
src_err_list[nerrs++] = i;
}
if (nerrs == 0){ // should have at least one error
while ((err = (rand() % KMAX)) >= k) ;
src_err_list[nerrs++] = err;
src_in_err[err] = 1;
}
// Construct b by removing error rows
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
for(j=0; j<k; j++)
b[k*i+j] = a[k*r+j];
}
if (gf_invert_matrix(b, d, k) < 0){
printf("BAD MATRIX\n");
return -1;
}
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
recov[i] = ubuffs[r];
}
for(i=0; i<nerrs; i++){
for(j=0; j<k; j++){
c[k*i+j]=d[k*src_err_list[i]+j];
}
}
// Recover data
ec_init_tables(k, nerrs, c, g_tbls);
ec_encode_data(size, k, nerrs, g_tbls, recov, &temp_ubuffs[k]);
for(i=0; i<nerrs; i++){
if (0 != memcmp(temp_ubuffs[k+i], ubuffs[src_err_list[i]], size)){
printf("Fail error recovery (%d, %d, %d) - ", m, k, nerrs);
printf(" - erase list = ");
for (i=0; i<nerrs; i++)
printf(" %d", src_err_list[i]);
printf("\na:\n");
dump_u8xu8((unsigned char*)a, m, k);
printf("inv b:\n");
dump_u8xu8((unsigned char*)d, k, k);
printf("orig data:\n");
dump_matrix(ubuffs, m, 25);
printf("orig :");
dump(ubuffs[src_err_list[i]],25);
printf("recov %d:",src_err_list[i]);
dump(temp_ubuffs[k+i], 25);
return -1;
}
}
// Confirm that padding around dests is unchanged
memset(temp_buffs[0], 0, PTR_ALIGN_CHK_B); // Make reference zero buff
for(i=0; i<m; i++){
offset = ubuffs[i] - buffs[i];
if (memcmp(buffs[i], temp_buffs[0], offset)){
printf("Fail rand ualign encode pad start\n");
return -1;
}
if (memcmp(buffs[i] + offset + size, temp_buffs[0], PTR_ALIGN_CHK_B - offset)){
printf("Fail rand ualign encode pad end\n");
return -1;
}
}
for(i=0; i<nerrs; i++){
offset = temp_ubuffs[k+i] - temp_buffs[k+i];
if (memcmp(temp_buffs[k+i], temp_buffs[0], offset)){
printf("Fail rand ualign decode pad start\n");
return -1;
}
if (memcmp(temp_buffs[k+i] + offset + size, temp_buffs[0], PTR_ALIGN_CHK_B - offset)){
printf("Fail rand ualign decode pad end\n");
return -1;
}
}
putchar('.');
}
// Test size alignment
align = (LEN_ALIGN_CHK_B != 0) ? 13 : 16;
for(size=TEST_LEN; size>0; size-=align){
while ((m = (rand() % MMAX)) < 2);
while ((k = (rand() % KMAX)) >= m || k < 1);
if (m>MMAX || k>KMAX)
continue;
for(i=0; i<k; i++)
for(j=0; j<size; j++)
buffs[i][j] = rand();
gf_gen_rs_matrix(a, m, k);
// Make parity vects
ec_init_tables(k, m-k, &a[k*k], g_tbls);
ec_encode_data_sse(size, k, m-k, g_tbls, buffs, &buffs[k]);
// Random errors
memset(src_in_err, 0, TEST_SOURCES);
for (i=0, nerrs=0; i<k && nerrs<m-k; i++){
err = 1 & rand();
src_in_err[i] = err;
if (err)
src_err_list[nerrs++] = i;
}
if (nerrs == 0){ // should have at least one error
while ((err = (rand() % KMAX)) >= k) ;
src_err_list[nerrs++] = err;
src_in_err[err] = 1;
}
// Construct b by removing error rows
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
for(j=0; j<k; j++)
b[k*i+j] = a[k*r+j];
}
if (gf_invert_matrix(b, d, k) < 0){
printf("BAD MATRIX\n");
return -1;
}
for(i=0, r=0; i<k; i++, r++){
while (src_in_err[r])
r++;
recov[i] = buffs[r];
}
for(i=0; i<nerrs; i++){
for(j=0; j<k; j++){
c[k*i+j]=d[k*src_err_list[i]+j];
}
}
// Recover data
ec_init_tables(k, nerrs, c, g_tbls);
ec_encode_data(size, k, nerrs, g_tbls, recov, &temp_buffs[k]);
for(i=0; i<nerrs; i++){
if (0 != memcmp(temp_buffs[k+i], buffs[src_err_list[i]], size)){
printf("Fail error recovery (%d, %d, %d) - ", m, k, nerrs);
printf(" - erase list = ");
for (i=0; i<nerrs; i++)
printf(" %d", src_err_list[i]);
printf("\na:\n");
dump_u8xu8((unsigned char*)a, m, k);
printf("inv b:\n");
dump_u8xu8((unsigned char*)d, k, k);
printf("orig data:\n");
dump_matrix(buffs, m, 25);
printf("orig :");
dump(buffs[src_err_list[i]],25);
printf("recov %d:",src_err_list[i]);
dump(temp_buffs[k+i], 25);
return -1;
}
}
}
printf("done EC tests: Pass\n");
return 0;
}
